Instruction on IFEval (avg@32 accuracy)
46.2Avg@32 AccuracyIBTPO
Evaluation Results
| Method | Links | |
|---|---|---|
| IBTPOBase Model=Qwen3-8B-Base2026.05 | 46.2 | |
| GRPO w/ Clip-higherBase Model=Qwen3-8B-Base2026.05 | 45.4 | |
| IBROBase Model=Qwen3-8B-Base2026.05 | 44.7 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-8B-Base2026.05 | 44.2 | |
| Initial ModelBase Model=Qwen3-8B-Base2026.05 | 43.8 | |
| TreeRLBase Model=Qwen3-8B-Base2026.05 | 42.5 | |
| TreePOBase Model=Qwen3-8B-Base2026.05 | 42.2 | |
| Vanilla GRPOBase Model=Qwen3-8B-Base2026.05 | 42 | |
| IBTPOBase Model=Qwen3-1.7B-Base2026.05 | 26.9 | |
| IBROBase Model=Qwen3-1.7B-Base2026.05 | 25.4 | |
| Vanilla GRPOBase Model=Qwen3-1.7B-Base2026.05 | 24 | |
| GRPO w/ Clip-higherBase Model=Qwen3-1.7B-Base2026.05 | 23.7 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-1.7B-Base2026.05 | 23.7 | |
| TreeRLBase Model=Qwen3-1.7B-Base2026.05 | 23.5 | |
| TreePOBase Model=Qwen3-1.7B-Base2026.05 | 23.2 | |
| Initial ModelBase Model=Qwen3-1.7B-Base2026.05 | 21.9 |